Neuro-wideband WiFi Sensing Enables Multipath Resolution Beyond Limited Bandwidths

The limitations of current WiFi sensing technology, hampered by narrow bandwidths and resulting in poor multipath resolution, are now being addressed by innovative new research. Sijie Ji, Weiying Hou from The University of Hong Kong, and Chenshu Wu present Neuro-Wideband (NWB), a novel approach to wideband WiFi sensing that requires no specialised hardware or additional channel measurements. This work centres on the idea that existing channel state information (CSI) contains inherent multipath parameters which, when processed correctly, can be extrapolated to approximate measurements across a much broader bandwidth. By introducing WUKONG, a learning framework combining Transformer and Diffusion models, the researchers demonstrate a pathway to unlock the potential of commodity WiFi hardware for applications like precise localisation and even remote breathing monitoring, significantly expanding the possibilities for wireless sensing systems.

Exploiting CSI for Wideband WiFi Sensing

A significant limitation in multipath resolution and multi-user sensing is the difficulty of obtaining large bandwidths on commercial WiFi networks, stemming from limited spectrum availability and increasingly crowded conditions. This paper introduces Neuro-Wideband (NWB), a novel paradigm to facilitate wideband WiFi sensing without specialised hardware or additional channel measurements. The central concept is that any physical measurement of channel state information (CSI) intrinsically contains multipath parameters. These parameters, while individually unsolvable, can be transformed into an expanded form of CSI, termed eCSI, which effectively approximates measurements across a broader bandwidth. By leveraging this transformation, NWB overcomes the bandwidth limitations of conventional WiFi sensing techniques, enabling high-resolution sensing applications utilising existing WiFi infrastructure.

Neuro-Wideband CSI Reconstruction via Deep Learning

Researchers pioneered Neuro-Wideband (NWB) to extrapolate wideband Channel State Information (CSI) from standard narrowband measurements, circumventing the need for specialized hardware or additional channel probing. The approach leverages the principle that underlying multipath parameters remain constant across frequencies at a given time and location, leaving detectable signatures within any frequency’s CSI reading. To realise NWB, scientists developed WUKONG, a deep learning system designed to generate high-fidelity extended CSI (eCSI) from narrowband inputs. WUKONG was formulated as a self-conditioned learning problem, utilising CSI at 20, 40, and 80MHz to predict CSI at larger target bandwidths.

This ensures the resulting eCSI remains physically meaningful by grounding predictions in real, measured data, eliminating the requirement for wideband ground-truth labels during training. A key innovation within WUKONG lies in its ability to infer sample-specific multipath structure, beyond typical deep learning methods focusing on population-level statistics. Scientists integrated a frequency-aware Transformer with self-conditioned diffusion modelling, enabling the system to internalise multipath patterns and transfer this knowledge across frequency bands. Experiments employed both publicly available and self-collected data, demonstrating WUKONG’s capacity to handle arbitrary CSI lengths and bandwidths, generalise across hardware and protocols, and produce continuous eCSI even in dynamic environments. The research demonstrated high-fidelity eCSI extrapolation, notably from 20MHz to 160MHz, with strong alignment to measured 160MHz CSI. Case studies focusing on localization and multi-user breathing estimation validated the utility of NWB sensing in real-world applications, establishing a pathway towards ultrawide-band sensing using commodity WiFi hardware.

Wideband WiFi Sensing From Narrowband Signals

Scientists have developed Neuro-Wideband (NWB), a novel paradigm for wideband WiFi sensing that operates without requiring specialized hardware or additional channel measurements. The research addresses the restricted bandwidths traditionally allocated for communication, which constrain multipath resolution. NWB successfully extrapolates wideband Channel State Information (eCSI) from a single narrowband measurement, expanding the available bandwidth for sensing applications and improving range resolution. The team proposes WUKONG, a deep learning framework designed to realize the NWB paradigm, formulated as a unique self-conditioned learning problem.

WUKONG utilizes existing CSI data as self-labeled samples, eliminating the need for costly wideband ground truth labels. Measurements confirm that WUKONG integrates a Transformer and Diffusion model, capturing sample-specific multipath parameters and transferring this knowledge to generate the expanded eCSI, allowing the system to adapt to complex, dynamic environments. Real-world experiments demonstrate the effectiveness of NWB across diverse WiFi signals and protocols. Data shows that WUKONG generates high-fidelity eCSI, enabling applications such as localization and multi-person breathing monitoring. Campus WiFi network measurements indicated that less than 1% of access points currently operate at the maximum 160MHz bandwidth, with most devices functioning within the 20, 80MHz range. This work overcomes limitations of existing methods, such as frequency hopping, by providing a continuous and physically meaningful wideband CSI signal.

WUKONG extrapolates wideband CSI for sensing applications Researchers

This work introduces Neuro-Wideband (NWB), a new approach to wideband WiFi sensing that does not require specialised hardware or additional channel measurements. Researchers developed WUKONG, a deep learning framework capable of extrapolating existing channel state information (CSI) into an expanded continuous-bandwidth representation termed eCSI. By transferring knowledge of sample-specific multipath characteristics, WUKONG generates high-fidelity eCSI data, effectively simulating a wider bandwidth than is physically available. Experiments demonstrate WUKONG’s ability to accurately extrapolate eCSI, validated physically up to 160MHz, and its robustness across different environments and hardware. Case studies focusing on localisation and multi-person breathing monitoring illustrate the practical benefits of utilising eCSI for sensing applications. Future research should explore applying NWB and WUKONG to more complex sensing scenarios and refining the extrapolation process, representing a significant step towards enabling high-resolution, real-time sensing capabilities on standard wireless systems.